Mismatch for the depth map checkpoint from MiDaS.

[yuzhou914]

Hello, thanks for your great work. When I use the training code with depth condition, I download the depth checkpoint from dpt_swin2_large_384, the official checkpoint given by MiDaS. However, I meet the mismatch problem when I load the checkpoint here. It is strange, do you know what is the problem?

RuntimeError: Error(s) in loading state_dict for DPTDepthModel: Missing key(s) in state_dict: "pretrained.model.layers.3.downsample.reduction.weight", "pretrained.model.layers.3.downsample.norm.weight", "pretrained.model.layers.3.downsample.norm.bias", "pretrained.model.head.fc.weight", "pretrained.model.head.fc.bias". Unexpected key(s) in state_dict: "pretrained.model.layers.0.downsample.reduction.weight", "pretrained.model.layers.0.downsample.norm.weight", "pretrained.model.layers.0.downsample.norm.bias", "pretrained.model.layers.0.blocks.1.attn_mask", "pretrained.model.layers.1.blocks.1.attn_mask", "pretrained.model.head.weight", "pretrained.model.head.bias". size mismatch for pretrained.model.layers.1.downsample.reduction.weight: copying a param with shape torch.Size([768, 1536]) from checkpoint, the shape in current model is torch.Size([384, 768]). size mismatch for pretrained.model.layers.1.downsample.norm.weight: copying a param with shape torch.Size([768]) from checkpoint, the shape in current model is torch.Size([384]). size mismatch for pretrained.model.layers.1.downsample.norm.bias: copying a param with shape torch.Size([768]) from checkpoint, the shape in current model is torch.Size([384]). size mismatch for pretrained.model.layers.2.downsample.reduction.weight: copying a param with shape torch.Size([1536, 3072]) from checkpoint, the shape in current model is torch.Size([768, 1536]). size mismatch for pretrained.model.layers.2.downsample.norm.weight: copying a param with shape torch.Size([1536]) from checkpoint, the shape in current model is torch.Size([768]). size mismatch for pretrained.model.layers.2.downsample.norm.bias: copying a param with shape torch.Size([1536]) from checkpoint, the shape in current model is torch.Size([768]).

[HL-hanlin]

Hi Yuzhou,

Thanks for trying our model!

I remembered similar issue happened to me because timm library version is too high. In requirement_train.txt, we install timm==0.6.12. Could you try to downgrade your timm library version to it?

You can also see a reference link about this issue here

Let me know if this helps or not!

[yuzhou914]

Hello, even though I reinstall the timm==0.6.12, I will still meet the problem when I use: helper.add_depth_estimator(estimator_ckpt_path = os.path.join(DATA_PATH, "ckpts/DepthMidas/dpt_swin2_large_384.pt")). Do you have any other mind? Thanks.

[HL-hanlin]

I don't have other better ideas for now lol.

Maybe could you double check if the timm version is correct via command "pip list"? If the problem still exist, you can let me know your email/wechat, and I'll set up a zoom meeting so that we can debug together. Thanks!